The health data analyst

In one line

A health data analyst turns clinical and operational data into decisions. The technical skills — SQL, statistics, a visualisation tool — are the entry ticket. The job is knowing why health data lies, and being the person who says so.

Why health data is not just data

Any competent analyst can group and count. Health data punishes that instinct specifically, because almost every field means something other than what it appears to mean.

Billing codes are not clinical facts. A diagnosis on a claim exists to justify payment. It may be the most reimbursable of several true diagnoses, or a "rule-out" that was never confirmed. Count ICD-10-CM codes and call it disease prevalence and you have measured billing behaviour — confidently, precisely, and wrongly.

Absence is ambiguous. A missing lab result can mean the test was normal and not repeated, never ordered, ordered elsewhere, or that the patient died. Those are four different worlds and the NULL looks identical in all of them.

The denominator moves. "Readmission rate went up" might mean care got worse — or that a competing hospital closed, or that coding got more thorough, or that the patient mix aged.

Time is not what it seems. The timestamp is usually when someone typed, not when the event happened. Overnight events land at 8am when the day team documents them.

The analyst who doesn't know this produces beautiful, fast, wrong dashboards. Producing them faster is not an improvement.

What the work actually looks like

• SQL, deeply. Not "I can SELECT" — joins across messy clinical schemas, window functions over longitudinal records, and the patience to find why the row count changed.
• A statistical conscience. You don't need to be a biostatistician, but you must know when you've left your competence. Confounding, regression to the mean, and the difference between a real signal and a small denominator will all bite you.
• Clinical literacy. Enough to know an eGFR of 26 is serious, that a "normal" range differs by lab and by age, and when to walk to a clinician and ask.
• Data models. OMOP CDM if you touch research or real-world evidence — it exists precisely so analyses can be compared across institutions.
• Communication. The finding that never reached a decision-maker in a form they trusted did not happen.

The line that defines the job

Here is the thing worth carrying: your output changes what someone does to a patient.

A retail analyst who overstates a trend costs the company money. A health analyst who overstates a trend can cause a service to be cut, a protocol changed, or a population deemed low-risk when it isn't. That asymmetry is why the defining professional skill is not technique but calibrated honesty — reporting the uncertainty as clearly as the number, and refusing to let a caveat get dropped between your notebook and the slide.

The most valuable sentence in this career is "I can show you that, but here's why I wouldn't trust it yet."

Getting in

The realistic entry paths, in rough order of how often they actually work:

• From a clinical role. Nurses, coders and pharmacists start with the thing that's hardest to teach — knowing what the data means — and add SQL. This is a genuinely strong route and it's undersold.
• From general analytics. You bring technique and must earn the clinical literacy. Faster to start, slower to become trustworthy.
• From RCM or coding. You already know exactly why billing codes lie. That's not a small head start; it's the thing most analysts learn last, expensively.

Build something real and show your reasoning: the analytics and data-science labs here work on synthetic patients precisely so you can practise the judgement without risking anyone's records. A portfolio that shows you found a confound is worth more than one that shows you made a chart.